Working with data in R

table() is useful for counting the number of observations in different groups or levels

color <- c("red", "red", "red", "blond", "blond", "blond", "blond", "brunette", "brunette", "brunette", "brunette")
style <- c("perm", "mullet", "perm", "mohawk", "mullet", "mullet", "braids", "mullet", "mullet", "perm", "braids")
 
table(color, style)
#          style
# color    braids mohawk mullet perm
# blond         1      1      2    0
# brunette      1      0      2    1
# red           0      0      1    2

tapply() is very useful for applying a function (e.g., mean, max, length, or one of your own) to the observations in different groups or levels

length.in.back <- c(2, 5, 1.5, 8, 3, 4.5, 13, 9, 2, 3.5, 3)
 
tapply(length.in.back, INDEX = style, FUN = mean)
#   braids   mohawk   mullet     perm
# 8.000000 8.000000 4.700000 2.333333
 
tapply(length.in.back, INDEX = style, FUN = function(x) { sd(x)/mean(x) }) #calculating the C.V.
#    braids mohawk    mullet      perm
# 0.8838835     NA 0.5709110 0.4460713

You can find a nice example of how to use tapply() and other variants of apply() here: http://nsaunders.wordpress.com/2010/08/20/a-brief-introduction-to-apply-in-r/

melt() and dcast() are useful functions when you're trying to reshape you're data, going from wide to long or vice versa. It's part of the reshape package.
For wide to long:

library(reshape)
wide <- read.csv("IENFD.csv")
 
#  ID IENFD.DL IENFD.DT IENFD.PT Sex
#1  1     2.10     5.42    11.70   M
#2  2     6.63     7.94     8.58   M
#3  3     9.56    14.21     9.10   F
#4  4     6.18     6.46     5.52   F
 
#with melt you provide the ID variables (i.e., those that were not measured) and it assumes everything else was measured.
 
long <- melt(wide, id.vars = c("ID", "Sex"))
 
#the output is a new dataframe with fewer columns and more rows
#in this case, "ID" and "Sex" were kept as the index variables, and R created new columns "variable" and "value"
 
#   ID Sex variable value
#1   1   M IENFD.DL  2.10
#2   2   M IENFD.DL  6.63
#3   3   F IENFD.DL  9.56
#4   4   F IENFD.DL  6.18
#5   1   M IENFD.DT  5.42
#6   2   M IENFD.DT  7.94
#7   3   F IENFD.DT 14.21
#8   4   F IENFD.DT  6.46
#9   1   M IENFD.PT 11.70
#10  2   M IENFD.PT  8.58
#11  3   F IENFD.PT  9.10
#12  4   F IENFD.PT  5.52
 
#the "variable" column is the ID of your repeated measure, and the "value" is the corresponding measures
#you can rename the columns with the code:
 
colnames(long)[c(3,4)] <- c("Location", "IENFD")
 
#   ID Sex variable value
#1   1   M IENFD.DL  2.10
#2   2   M IENFD.DL  6.63
#3   3   F IENFD.DL  9.56
 
#You can also clean up the "variable" names by changing the labels on the levels:
 
levels(long$Location)
 [1] "IENFD.DL" "IENFD.DT" "IENFD.PT"
levels(long$Location) <- c("DL", "DT", "PT")
 
#   ID Sex Location IENFD
#1   1   M       DL  2.10
#2   2   M       DL  6.63
#3   3   F       DL  9.56
#4   4   F       DL  6.18
#5   1   M       DT  5.42
#6   2   M       DT  7.94
#7   3   F       DT 14.21
#8   4   F       DT  6.46
#9   1   M       PT 11.70
#10  2   M       PT  8.58
#11  3   F       PT  9.10
#12  4   F       PT  5.52
 

For more help with reshaping data, go here: http://wiki.stdout.org/rcookbook/Manipulating%20data/Converting%20data%20between%20wide%20and%20long%20format/

model.matrix() When you have factors as predictor variables, it can be very useful to generate a design matrix that specifies which effects (beta parameters) should come into play for each data point. The design matrix is essentially your data, re-formatted as a matrix of zeros and ones specifying the categorical attributes of each data point.In R, design matrices are easily generated using the model.matrix function.
For example, say you had a large dataset with a categorical predictor variable with 10 levels, for which it would otherwise be tedious to assemble a design matrix (e.g., 10 different species).
The simple R command x <- model.matrix(~species) will automatically generate the design matrix for modeling the effect of species with an intercept term (use model.matrix(~species-1) if you don't want an intercept term). If you had 2 categorical variables you wanted to include as predictor variables (e.g., species and habitat type), you could generate a design matrix for a model including main and interaction effects with the R command x <- model.matrix(~species*habtype).

For the seed predation example from Bolker's chapter 8, I used model.matrix to specify a model for the effect of species on seed predation rates as follows:

species_ind <- model.matrix(~species-1) # design matrix for species- no intercept
params <- numeric(9)
 
NLL_sp_species3 <- function(params, numeaten, available) {
   ltheta <- params[1]
   betas <- c(params[2:9])
   -sum(dbetabinom(x=numeaten,size=available,prob=plogis(species_ind %*% betas),
   theta=exp(ltheta),log=TRUE))
}
 
SP.species3 <- optim(fn=NLL_sp_species3, par=c(ltheta=log(0.35),
  betas=rep(qlogis(0.05),times=8)),
  numeaten = taken,
  available = available,
  method="BFGS")